原型 · 预览版:收录官方注册中心里的全部服务器,每 30 分钟同步一次。使用 Google 登录即可撰写评价。原型 · 每 30 分钟更新
PopNx MCP为使用 AI 的人准备的 MCP 服务器 · 用大白话讲清楚

tc39-mcp

作者:xyzzylabs版本 0.6.6i★ 2i电脑或云端官方网站 ↗
注册名称io.github.xyzzylabs/tc39-mcp

安装方法

What it does

Independent MCP server for the TC39 specs (ECMA-262 + ECMA-402): clauses, search, diffs, history.

  • Connects 电脑或云端
  • Account or key No key needed
  • Category 搜索与 AI 记忆

可搭配使用

教 AI 做这类工作的 Skill,可与本项一起使用

timesfm

TimesFM (Time Series Foundation Model) is a pretrained time-series foundation model developed by Google Research for time-series forecasting.

作者:google-research★ 35,6291 个 skill

Skill · 操作手册

Shadowbroker

Open-source intelligence for the global theater. Track everything from the corporate/private jets of the wealthy, and spy satellites, to seismic events in one unified interface. Hook an AI agent up to have it parse through data and find previously unseen correlations. The knowledge is available to…

作者:BigBodyCobain★ 11,6021 个 skill

Skill · 操作手册

google-maps-scraper

scrape data from Google Maps. Extracts data such as the name, address, phone number, website URL, rating, reviews number, latitude and longitude, reviews,email and more for each place

作者:gosom★ 8,2091 个 skill

Skill · 操作手册

为什么排在这里

此项在“连接 AI 的应用”标签中,该标签不使用星标,按以下顺序排列

1. 在 Claude 连接器目录中否
2. 团队实测过否
3. 适合泰国用户否
4. 不计星标的使用量20/100热度 0 · 新鲜度 12 · 仍在响应 8 · 用户喜欢 +0
5. 登记日期2026-11-07

用户喜欢: 0 · 最多加 5 分(每人 1 分),不能用钱买。

质量: 基础检查通过 5/5 项(仅展示,不用于排序)

安装

选择你使用的 AI 应用,然后复制设置。

用一句话让 AI 帮你安装

请帮我安装 tc39-mcp。按照 https://mcp.popnx.com/i/xyzzylabs-tc39-mcp.md 操作,先把命令给我看,等我同意后再运行。

粘贴到能在你电脑上工作的 AI(如 Claude Code、Cursor)。它会读取我们的步骤,先给你看命令,等你同意。无需密钥,无需注册。 · 查看 AI 会读取的全文

通过 5/5 项检查 · 你的提问会经过开发者的服务器(mcp.xyzzylabs.ai) · ✓ 命令已检查 2026年11月13日

第一步:你用哪个 AI?

选择你用的应用,步骤会显示在下方。

可连接 Claude · 约 2 分钟

不需要密钥,不用安装程序 · 在 claude.ai 或桌面应用设置一次,即可在聊天中使用 · 免费账户可添加 1 个 · Team/Enterprise 方案由所有者(Owner)添加

  1. 打开 Claude 的 Connectors 页面

    CustomizeConnectors

    CustomizeConnectors1Add2https://…3Add4
  2. 点击“Add” → “Custom” → “Web”,填写这 2 项

    把这里的内容复制到 Claude 中同名的输入框,然后点击 Continue

    Name · 名称tc39-mcp
    Remote MCP server URL · 地址https://mcp.xyzzylabs.ai/tc39/mcp

    在 Claude 检测到的登录设置页点击 Continue · Authentication 按作者说明选择(Sign in now、Sign in when needed 或 No sign in),然后点击 Add

  3. 在聊天中开启

    +Connectorstc39-mcp

  4. 试一试

    用 tc39-mcp 告诉我你能帮我做什么,然后做一件给我看

    ✓ 如果 Claude 说正在使用 tc39-mcp,或回答中出现 tc39-mcp 的工具名称,就表示连接成功

菜单名称已于 2026年11月11日 对照 Anthropic 帮助页面核实 · 你的界面可能因版本略有不同

能否连接 ChatGPT 取决于方案

完整 MCP:网页版 ChatGPT Business、Enterprise 或 Edu · Pro:以读取/获取权限连接,修改数据的工具可能用不了 · Plus:官方文档尚未确认,如果在插件页看到添加自定义 MCP 的选项可以试试 · 仅限网页

  1. 在网页版 ChatGPT 打开添加 MCP 的页面

    Plugins+Add custom MCP server

    ChatGPTPlugins1+2https://…3Create4
  2. 填写名称和地址

    输入名称,在 Connection 下粘贴地址,勾选风险提示,然后点击 Create as a plugin

    Name · 名称tc39-mcp
    Connection · 地址https://mcp.xyzzylabs.ai/tc39/mcp
  3. 在聊天中选择 tc39-mcp

    开启新聊天,输入前先从插件列表中选择 tc39-mcp

  4. 试一试

    用 tc39-mcp 告诉我你能帮我做什么,然后做一件给我看

    ✓ 如果 ChatGPT 说正在使用 tc39-mcp,或回答中出现 tc39-mcp 的工具名称,就表示连接成功

菜单名称已于 2026年11月11日 对照 OpenAI 帮助页面核实 · 你的界面可能因版本略有不同

可连接 Claude Code · 约 2 分钟

不需要密钥,不用安装程序

  1. 打开终端

    Mac:按 Command+空格,输入 Terminal · Windows:打开 PowerShell

  2. 粘贴这条命令并按 Enter

    claude mcp add --transport http --scope user tc39 https://mcp.xyzzylabs.ai/tc39/mcp
  3. 在 Claude Code 中输入 /mcp

    应能看到 tc39-mcp 在列表中并显示已连接

  4. 试一试

    用 tc39-mcp 告诉我你能帮我做什么,然后做一件给我看

    ✓ 如果 Claude Code 说正在使用 tc39-mcp,或回答中出现 tc39-mcp 的工具名称,就表示连接成功

菜单名称已于 2026年11月11日 对照 Anthropic 帮助页面核实 · 你的界面可能因版本略有不同

可连接 Cursor · 约 2 分钟

不需要密钥,不用安装程序

  1. 点击这个按钮打开 Cursor

    在 Cursor 中打开

    Cursor 会请你确认安装,点击 Install

  2. 在 Cursor 中以 Agent 模式打开 AI 聊天

  3. 试一试

    用 tc39-mcp 告诉我你能帮我做什么,然后做一件给我看

    ✓ 如果 Cursor 说正在使用 tc39-mcp,或回答中出现 tc39-mcp 的工具名称,就表示连接成功

菜单名称已于 2026年11月11日 对照 Cursor 帮助页面核实 · 你的界面可能因版本略有不同

可连接 VS Code · 约 2 分钟

不需要密钥,不用安装程序

  1. 点击这个按钮打开 VS Code

    在 VS Code 中打开

    VS Code 会请你确认安装,点击 Install

  2. 在 VS Code 中以 Agent 模式打开 AI 聊天

  3. 试一试

    用 tc39-mcp 告诉我你能帮我做什么,然后做一件给我看

    ✓ 如果 VS Code 说正在使用 tc39-mcp,或回答中出现 tc39-mcp 的工具名称,就表示连接成功

菜单名称已于 2026年11月11日 对照 VS Code 帮助页面核实 · 你的界面可能因版本略有不同

其他支持 MCP 的应用(如 Windsurf、Zed、Gemini CLI)添加方式各不相同,请在应用列表中查看

查看所有支持 MCP 的应用 →

通过网页连接的应用使用的地址

地址https://mcp.xyzzylabs.ai/tc39/mcp

JSON 设置在下方“开发者专用”中

卡住了?问问正在用 tc39-mcp 的人💬 讨论 →
开发者专用:终端命令和 JSON 设置

通过网页连接

Claude Code

claude mcp add --transport http --scope user tc39 https://mcp.xyzzylabs.ai/tc39/mcp

Cursor (~/.cursor/mcp.json)

{
  "mcpServers": {
    "tc39": {
      "url": "https://mcp.xyzzylabs.ai/tc39/mcp"
    }
  }
}

VS Code (mcp.json)

{
  "servers": {
    "tc39": {
      "type": "http",
      "url": "https://mcp.xyzzylabs.ai/tc39/mcp"
    }
  }
}

安装在电脑上

Claude Code

claude mcp add --scope user tc39 -- npx -y tc39-mcp

Claude Desktop (claude_desktop_config.json)

{
  "mcpServers": {
    "tc39": {
      "command": "npx",
      "args": [
        "-y",
        "tc39-mcp"
      ]
    }
  }
}

Cursor (~/.cursor/mcp.json)

{
  "mcpServers": {
    "tc39": {
      "command": "npx",
      "args": [
        "-y",
        "tc39-mcp"
      ]
    }
  }
}

VS Code (mcp.json)

{
  "servers": {
    "tc39": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "tc39-mcp"
      ]
    }
  }
}

这些设置是根据作者在注册中心的条目生成的。如果 README 的写法不同,请以 README 为准。

一览

每周下载量i
98/周
连接方式
电脑或云端
传输方式
stdio, streamable-http
最新版本i
0.6.6
发布到注册中心i
2026年11月8日
代码最后更新i
2026年11月17日
最近检查
2026年11月18日
作者
xyzzylabs
软件包
npm: tc39-mcp
注册状态
活跃

哪些应用可以使用它

依据这个服务器的连接方式和各应用自己的文档,检查于 2026年11月10日。

通过互联网连接,适用于

ClaudeChatGPTGooseLM StudioMsty StudioCherry StudioCursorVS Code + GitHub CopilotVisual StudioWindsurf (Devin Desktop)ZedKiroClineContinueJetBrains AI AssistantClaude CodeCodex CLIGemini CLIOpenCodenodecodeWarpLibreChatOpen WebUIRaycast

在你的电脑上运行,适用于

ClaudeGooseLM StudioMsty StudioCherry StudioCursorVS Code + GitHub CopilotVisual StudioWindsurf (Devin Desktop)ZedKiroClineContinueJetBrains AI AssistantClaude CodeCodex CLIGemini CLIOpenCodenodecodeWarpLibreChatRaycast

全部应用及其优缺点 →

它能做什么

工具名称和描述与服务器在 2026年11月18日 报告的完全一致。这些内容由作者撰写,不是我们写的;我们以纯文本形式显示。

  • spec.about

    Self-description of this MCP server: package name + version, per-snapshot pin metadata (sha, fetched_at, biblio_commit, clause_count) for every supported (spec, edition), plus test262 + proposals index headers when present.

  • clause.get

    Fetch a parsed TC39 clause as structured JSON. `spec` selects '262' (default) or '402'. `edition` defaults to 'latest'. `at: ' '` pins to a historical main snapshot (only valid for edition='main'); omit to query the live snapshot.

  • clause.list

    List parsed spec clauses with optional filters (kind, section prefix, has_algorithm). `spec` selects '262' or '402'. `at: ' '` queries a historical main snapshot.

  • spec.search

    Search the parsed spec by clause id / aoid / title (+ step text with `search_steps`). Aoid-exact ranks first. `at: ' '` searches a historical main snapshot.

  • proposal.list

    List TC39 proposals from the static index (tc39/proposals). Filter by spec ('262'|'402'), stage ('0'|'1'|'2'|'2.7'|'3'|'finished'|'inactive'|'active'), champion (substring), or contains (name/slug substring).

  • proposal.get

    Fetch one TC39 proposal by slug (exact) or name (case-insensitive).

  • spec.grammar

    Query grammar productions captured from the spec's ` ` blocks. `{ nonterminal }` returns every production for that non-terminal (exact match); `{ contains }` filters by RHS / name substring; neither lists all non-terminals + their production counts. `include_sdo` folds in SDO-att

  • spec.tables

    List or fetch parsed ` ` content. `{ id }` returns exactly that table (full columns + rows); otherwise list table summaries, optionally narrowed by a `filter` substring over the caption or id.

  • spec.sdo_index

    Index Syntax-Directed Operations by the grammar production they're defined on. `by: 'production'` (default) groups SDOs under each production; `by: 'sdo'` groups productions under each SDO title. `filter` narrows to keys containing a substring (case-insensitive).

  • clause.outline

    Return the section tree (table of contents) for a parsed (spec, edition). `depth` caps how deep the tree descends (1 = top-level only); `under` limits the tree to descendants of one clause id.

  • spec.global_search

    Run spec.search across both ECMA-262 and ECMA-402 in one call and interleave hits by score. Each hit is tagged with the spec it came from. Use it when you don't know which spec defines a symbol. `search_steps` also matches algorithm step text.

  • spec.snapshots

    List the live (spec, edition, sha, fetched_at) snapshots the hosted Worker is serving from R2. Filter by `spec` ('262'|'402') or `edition` (e.g. 'main', 'es2026'). Historical SHA-pinned copies are reachable via `at:` on clause.get / spec.search but aren't enumerated here.

  • spec.symbol_resolve

    Resolve spec notation like `[[Prototype]]` (internal slot), `%Object.prototype%` (well-known intrinsic), or `~number~` (sigil enum): return clauses that mention or define it, ranked by occurrence with a bump for definition-y sections.

  • spec.well_known_intrinsics

    Enumerate the well-known intrinsics in a spec with their probable defining clauses. Driven from the canonical §6.1.7.4 WKI table when present (ECMA-262), else a `%X%` prose scan (e.g. ECMA-402). `filter` narrows by bare-name substring.

  • spec.diff

    Clause-level diff of one clause across two editions of a spec. Reports identical / modified / added / removed plus a field-level breakdown (title, signature, step count, reworded step indices, notes, crossrefs). `from` defaults to the latest stable release, `to` to main.

  • spec.crossrefs

    For a clause id, return its outgoing references (clauses it cites) and/or incoming references (clauses that cite it — the back-reference index the parse alone doesn't expose). The reverse index is AOID-densified from step text. `direction`: 'in' | 'out' | 'both' (default). Outgoi

  • test262.search

    Search the tc39/test262 conformance suite from its indexed front-matter. `query` AND-matches whitespace tokens (case-insensitive) across each test's description + path; `esid` prefix-matches the front-matter esid. Returns ranked hits (path, GitHub url at the indexed SHA, esid, de

基础检查i

我们检查的是任何人都能看到的信息;我们没有阅读代码。不适用的检查项不计入总数。 · 检查于 2026年11月19日

  • ✓ 通过

    作者与仓库一致

    注册名称已通过 GitHub 账户 xyzzylabs 验证,该账户拥有此仓库

  • ✓ 通过

    仍在维护

    代码最后更新于前天

  • ✓ 通过

    有许可证

    使用 MIT

  • ✓ 通过

    端点有响应

    在 261 毫秒内响应(HTTP 200)

  • – 不适用

    使用 OAuth 登录

    无需登录

  • ✓ 通过

    工具描述中没有隐藏指令

    已读取 17 个工具描述,未发现针对 AI 的隐藏指令

云端地址最近 4 次检查中有 4 次响应 · 最近一次 261 ms · 检查于 2026年11月18日

已公开的漏洞(OSV.dev)

OSV.dev 中没有 tc39-mcp@0.6.6 的记录

只检查这个版本的包本身,不包括所有依赖 · 检查于 2026年11月15日

我们检查什么 → · JSON

作者:申请重新检查或报告信息错误
给作者:在 README 中加徽章

如果你是作者,可把这段代码贴到 README。徽章会链接回本页。

[![Featured on PopNx MCP](https://mcp.popnx.com/badge/xyzzylabs-tc39-mcp.svg)](https://mcp.popnx.com/en/servers/xyzzylabs-tc39-mcp)

讨论

提问、分享技巧或报告无法使用。按“有帮助”数量排序。

还没有帖子,欢迎第一个提问或分享。

此处内容由用户撰写,不代表 PopNx 团队。

登录后发帖

用户评价

还没有评价

登录后写评价

与“搜索与 AI 记忆”中的其他服务器比较i

  • 星标数多于此分类中其他 6394 个服务器的 82%
  • 发布到注册中心的时间比此分类中其他 6394 个服务器的 78% 更新
  • 基础检查结果等于或优于此分类中其他 6394 个服务器的 99%

与注册中心里这个分类中所有活跃的服务器比较。只有当发布者拥有该仓库时,星标数才计入。

常见问题

tc39-mcp 安全吗?

它在 5 项适用的基础检查中通过了 5 项。我们没有阅读代码。在添加密钥或令牌之前,请先看看这个服务器要求哪些权限,并使用只具备所需权限的密钥。

它需要哪些密钥或代码?

注册中心没有列出必需的密钥,不过有些服务器会在首次使用时要求你登录。

它在我的电脑上运行还是在云端?

两种方式都可以:安装在你的电脑上,或者连接云端版本。

它还在维护吗?

GitHub 仓库最后更新于前天,最新版本(0.6.6)于11天前进入注册中心。

哪些应用可以使用它?

支持 MCP 的应用,例如 Claude Desktop、Claude Code、Cursor 和 VS Code。它通过 URL 连接,因此也能用于接受远程 MCP 的应用,例如网页版 ChatGPT(Business 及以上为完整 MCP;Pro 为只读)。

有什么可以替代它?

同一分类中有:OpenSEO, Exa, Microsoft Learn MCP, Docs, COTAL actions (cotal.ai)

功能相近的其他选择

同一分类、其他作者、命令已通过检查。官方的排在前面,再按星标排序。

如何看卡片:↓ 每周下载量i★ GitHub 星标i三大支柱:评价 · 检查 · 使用情况i更新 = 最新版本发布到注册中心的时间i

OpenSEO

Keyword research, site audits, rank tracking and AI visibility for your AI agent.

作者:every-app广泛使用★ 2.8万5天前更新

在云端运行✓ everyapp.com

Exa

Connect AI agents to Exa for web search, content fetching, and multi-step research.

作者:exa-labs广泛使用★ 52422个月前更新

在云端运行✓ 服务方 · exa.aiPopNx 团队已实际试用

Microsoft Learn MCP

Official Microsoft Learn MCP Server – real-time, trusted docs & code samples for AI and LLMs.

作者:MicrosoftDocs广泛使用★ 199712个月前更新

在云端运行✓ 服务方 · microsoft.comPopNx 团队已实际试用

Docs

Search and read Empryo's documentation. Read-only, no auth, no local access.

作者:proxysoul正在增长★ 16262个月前更新

在云端运行✓ empryo.com
与相似项目对比
本项
tc39-mcp
OpenSEOExaMicrosoft Learn MCPDocsCOTAL actions (cotal.ai)
运行位置云端或安装在电脑上都可以在云端运行,无需安装在云端运行,无需安装在云端运行,无需安装在云端运行,无需安装在云端运行,无需安装
需要密钥否否否否否否
可在 Claude 中按 + 添加否是是是否否
基础检查5/55/52/44/44/44/4
许可证MITMITMITCC-BY-4.0NOASSERTIONApache-2.0
价格登记中未列出费用登记中未列出费用登记中未列出费用登记中未列出费用登记中未列出费用登记中未列出费用

作者的 README

打开 README

按原文从作者的仓库复制。PopNx MCP 没有审查或背书其内容。HTML、图片和脚本已被移除。链接会离开本站。

tc39-mcp

📖 Docs: mcp.xyzzylabs.ai/tc39 — Get started · Tools · Cookbook · Editions · Architecture · Hosting

Independent project — not an official Ecma

International or TC39 publication. Reads the publicly

published ECMAScript specs (ECMA-262 + ECMA-402).

Give MCP-speaking AI agents structural access to the JS spec. Any client that speaks the Model Context Protocol can call clause.get sec-tonumber and get back parsed JSON (algorithm steps as discrete arrays, cross-references as ids, signatures as typed values) instead of being handed a 4 MB spec.html to grep through. Tools cover ECMA-262 (the core language) and ECMA-402 (the Intl API): clauses, algorithm steps, cross-references both ways, edition diffs, upstream git history, test262 search, proposal lookup. Every response is SHA-pinned to a specific upstream commit so anything an agent cites stays reproducible.

Snapshots resolve through a local cache → hosted Worker → bundled fallback chain. The stdio transport (npx tc39-mcp) fetches each snapshot from the hosted Cloudflare Worker on a cold cache, writes it under ~/.cache/tc39-mcp/, and serves it from disk thereafter — revalidating only when the local copy is older than ~4 hours (a conditional If-None-Match request). The npm package also bundles the latest stable + main editions of both specs plus the test262 and proposals indexes; when the Worker is unreachable, those are served straight from the package (the offline fallback — not written to the cache). The hosted Worker is also the HTTP alternative when you want a shared network endpoint; its R2 data refreshes from upstream every ~4 hours.

Install + first call

Wire it into any MCP client — the stdio launch command is the same everywhere, only the config file differs:

{

"mcpServers": { "tc39": { "command": "npx", "args": ["tc39-mcp"] } } }

A global install works too — npm i -g tc39-mcp, then run tc39-mcp.

The first run downloads the npm package (latest stable + main editions plus the proposals and test262 indexes are bundled). The first call for a given snapshot fetches it from the hosted Worker and caches it locally; subsequent calls are served from disk, revalidated against the Worker only after the ~4-hour freshness window. If the Worker is unreachable, the bundled editions still answer offline. Then in your client:

use clause.get to read sec-tonumber and show me the steps

You should see structured JSON back:

{

"meta": { "id": "sec-tonumber", "aoid": "ToNumber", "title": "ToNumber ( argument )", "number": "7.1.4", "kind": "op" }, "signatureRaw": "ToNumber ( _argument_: an ECMAScript language value, ): either a normal completion containing a Number or a throw completion", "algorithms": [ { "steps": [ { "text": "If _argument_ is a Number, return _argument_." }, { "text": "If _argument_ is either *undefined* or a Symbol, throw a *TypeError* exception." }, { "text": "If _argument_ is *null*, return *+0* 𝔽 ." }, "..." ]} ], "crossrefs": ["sec-tonumber-applied-to-the-string-type", "..."] }

Five-minute walkthrough: docs/getting-started.md.

Hosted HTTP

Point your client at the hosted Cloudflare Worker instead of running a local subprocess — same MCP protocol, no install:

{

"mcpServers": { "tc39": { "type": "http", "url": "https://mcp.xyzzylabs.ai/tc39/mcp" } } }

Traffic is rate-limited to 30 req/min per IP.

What it's good at

  • Letting an agent reason about the spec without hallucinating. Structured JSON answers ground the model on real spec text: step numbering, cross-reference targets, signature shapes, edition deltas, conformance tests. Anything cited resolves to a specific clause id at a specific SHA — easy to verify, easy to reproduce.
  • Finding the clause you want from a hint. spec.search ranks AOID-exact matches first; spec.symbol_resolve decodes [[Prototype]] / %Object.prototype% / ~enumerate~.
  • Following references both ways. spec.crossrefs returns what a clause cites AND who cites it. AOID-densified so bare mentions in step text count, not just hrefs. include_cross_spec resolves 262 ↔ 402 hops. (Cookbook recipe 1.)
  • Comparing editions and tracking prose drift. spec.diff between any two editions back to ES2016; spec.history walks the upstream git log via pickaxe search. (Cookbook recipe 2.)
  • Finding test262 coverage for a clause. test262.search with prefix-matched esid: catches sec-tonumber AND sec-tonumber-applied-to-the-string-type in one call.
  • Mapping proposals to the spec. proposal.list / proposal.get from a structured index of tc39/proposals, covering both ECMA-262 and ECMA-402 (Intl) proposals — filter by spec. Refreshed on the same 4-hour cadence as the specs.
  • Local cache, bundled fallback (stdio). Once a snapshot is cached under ~/.cache/tc39-mcp/, tool calls are served from disk and only revalidated against the hosted Worker after the ~4-hour freshness window (a conditional If-None-Match request that carries the R2 object key, never a clause-id). Bundled editions answer offline when the Worker is unreachable. The hosted Worker is the HTTP alternative for shared / multi-tenant use.

Tools (19 across 5 namespaces)

GoalTool(s)
Verify what's being servedspec.about · spec.snapshots
Read a specific clauseclause.get
Find a clause from a name / symptomspec.search · spec.global_search
Resolve [[X]] / %X% / ~X~ notationspec.symbol_resolve
Browse / outlineclause.list · clause.outline
Compare editions / commit historyspec.diff · spec.history
Walk references (in + out)spec.crossrefs
Read structured tablesspec.tables
Inspect the grammarspec.grammar · spec.sdo_index
Enumerate well-known intrinsicsspec.well_known_intrinsics
Find conformance teststest262.search · test262.get
Look up a proposalproposal.list · proposal.get

Full reference (input schemas, output types, example calls per tool): [`docs/tools.md`](docs/tools.md) — auto-generated from the schemas so it never drifts.

Specs + editions

Every spec-reading tool accepts spec ("262" or "402", default "262") and edition (default "latest").

  • ECMA-262: es2016 – es2026, main. (ES5 / ES5.1 / ES6 have no upstream tags and aren't supported.)
  • ECMA-402: es2016 – es2026, main. (402 publishes each annual edition as an esYYYY branch rather than a tag; the fetch step resolves a branch or a tag the same way.)
  • Aliases: latest is spec-aware (each spec → its current stable release, es2026 today). draft / next → main on both.

Full table + how to add new releases: docs/editions.md.

Self-hosting snapshots

The stdio server fetches snapshots from the public hosted Worker at https://mcp.xyzzylabs.ai/tc39/r2/ (cache → Worker → bundled fallback), so on a strict-egress network it falls back to the bundled editions and can't reach the others. Override the base URL via TC39_MCP_BASE_URL to point at a private mirror — useful for strict-egress networks, air-gapped environments, or running against a self-hosted Worker:

TC39_MCP_BASE_URL=https://my-mirror.example.com npx tc39-mcp

The endpoint just needs to serve the same key structure (spec- - .json, test262-index.json, proposals-index.json) — a plain static file server works. If it returns ETags, the server revalidates with If-None-Match (cheap 304s); without them it just refetches the full object when a cached copy goes stale. To populate a mirror, run npm run parse against a local checkout (see below) and upload build/*.json to your bucket of choice.

The cache lives at $XDG_CACHE_HOME/tc39-mcp (or ~/.cache/tc39-mcp when XDG_CACHE_HOME is unset).

Build from source (contributors)

End users don't need this — the npm package and the hosted Worker are the supported surfaces above. This is for working on the server itself.

git clone https://github.com/xyzzylabs/tc39-mcp

cd tc39-mcp npm install npm run fetch-spec # ~2 min, ~150 MB — both specs at every supported edition npm run parse # spec.html → build/spec- - .json npm run fetch-test262 # optional, enables test262.* (~300 MB) npm run build-test262-index npm run fetch-proposals # optional, enables proposal.* (~50 MB) npm run build-proposals-index npm run mcp # start the stdio MCP server against your source

Point your MCP client at your local source instead of the published bin:

{

"mcpServers": { "tc39": { "type": "stdio", "command": "npm", "args": ["run", "--silent", "mcp"], "cwd": "/abs/path/to/tc39-mcp" } } }

--silent keeps npm's lifecycle banner off stdout, so the MCP

client receives a clean JSON-RPC stream.

Docs

Hosted at mcp.xyzzylabs.ai/tc39 — searchable, dark-mode-friendly, auto-rebuilt on every refresh so /snapshots always reflects the live SHAs.

In-repo (also browseable on GitHub):

  • docs/getting-started.md — install → wire → first call → verify. Five minutes.
  • docs/tools.md — every tool, every field, every example. Auto-generated from source.
  • docs/cookbook.md — multi-tool recipes: cross-spec lookups, prose-drift tracking, grammar/SDO cross-references, test262 coverage, proposal-to-clause mapping.
  • docs/editions.md — supported editions + alias resolution.
  • docs/architecture.md — data pipeline, parser, cache, memory model.
  • docs/deployment.md — local stdio, npm CLI, hosted Cloudflare Worker, refresh model, observability.
  • CONTRIBUTING.md — what kinds of changes land easily, what won't.
  • SECURITY.md — threat model + responsible disclosure.
  • CHANGELOG.md — version history + auto-refresh convention.

Privacy Policy

tc39-mcp is a read-only spec lookup service. The stdio transport (npx tc39-mcp) sends no telemetry and never transmits your queries — snapshots are fetched from the hosted Worker on a cold or stale cache and served from local disk otherwise; those fetches carry R2 object keys, never clause-ids or tool arguments. The hosted Cloudflare Worker collects only standard request metadata (IP for rate limiting, timestamps, request headers); it does not log request bodies, set cookies, or share data with third parties.

Full policy: mcp.xyzzylabs.ai/tc39/privacy

For privacy questions, open an issue with the privacy label on [GitHub](https://github.com/xyzzylabs/tc39-

只显示了开头部分;其余内容请在仓库中阅读